Vocco — On-Device Dictation & Speech to Text for Mac

Recording · on-device ⌘R

AirPods Pro Qwen

0:00

0:06

0:11

0:15

0:19

board-meeting.m4a42:10 · 58 MB

Transcribing · 21× realtime

Alex00:04

Mia00:11

Alex00:18

Most engines fall apart when you switch languages. Ours don't.

Real speech doesn't stay in one language — names, terms and whole clauses jump between languages mid-sentence. Vocco ships the engines that handle it, benchmarked in the open.

Mixed Chinese–English accuracy MER · ASCEND · N=300 · LOWER IS BETTER

Qwen3 ASRVocco default 0.152

Paraformer Large 0.176

SenseVoice Small 0.189

Whisper large-v3-turbo 0.277

All four run fully on-device in Vocco. Whisper large-v3-turbo is the engine most Mac dictation apps rely on.

Code-switching, not code-crashing

Switch languages mid-sentence — the transcript keeps up without a language picker.

Single languages, too

Pure English or pure Chinese dictation is just as strong — mixing is the hard case, not the only case.

Benchmarked in the open

Published test set, reproducible numbers — not a marketing adjective.

Punctuated & segmented

On-device voice detection splits your speech into clean, punctuated sentences as you talk.

Prompt your agent at speaking speed.

Long prompts are where typing hurts most. Hold the key, describe what you want — technical terms, file names, mixed languages and all — release, and the prompt is in the box.

Technical vocabulary survivesuseEffect, Kubernetes, PostgreSQL — jargon comes out spelled like jargon.
No time limit, no meterThink out loud for three minutes — local processing costs nothing per second.
Every prompt is keptLocal searchable history — copy any past dictation and re-send it to another agent.

WORKS IN CURSOR · CLAUDE CODE · WINDSURF · TERMINAL · ANY TEXT FIELD

Privacy isn't a promise. It's the architecture.

Transcription runs on your Mac's own silicon. Not "encrypted in transit" — there is no transit.

InputMicrophone

──▶

ProcessingYour Mac · local ASR

──▶

OutputYour cursor

Cloud ASR server ✕ This line doesn't exist. There is no endpoint to receive your audio.

Scenario 01

On a plane. Airplane mode on, Wi-Fi off — dictation works exactly the same.

Scenario 02

Under NDA. Client names, unreleased numbers — nothing to leak, because nothing leaves.

Scenario 03

Verify it yourself. Open Little Snitch while dictating. Watch the zero.

Hold. Speak. Release.

Step 01Hold right ⌘A quiet recording orb appears. Works in any app — Mail, Slack, your editor.

Step 02Speak naturallyText streams live as you talk, segmented into clean sentences by on-device voice detection.

Step 03Release — it's pastedThe text is already at your cursor, punctuated. Every dictation saved to local history.

File transcription

Drop in audio or video — transcript with per-segment timestamps in seconds.

Speaker diarization

Meeting recordings come back split by speaker. Models ship inside the app, offline.

SRT export

Every file transcript exports to SRT for subtitles, or stays as searchable notes.

Five engines

Start with Apple Speech; add Qwen3, SenseVoice, Paraformer or Whisper in one click.

A coffee a month. Or own it forever.

Cloud dictation bills $12–15 a month because every word costs them server time. Vocco runs on your own hardware — pick whichever suits you.

Monthly

$6.99/ month

billed monthly · cancel anytime

✓All features · all your Macs ✓Updates included while subscribed ✓Cancel anytime, runs to period end ✓7-day free trial, no card

Download & try free

Lifetime

$69.99once

one payment · yours forever

✓All features · all your Macs ✓Updates included, forever ✓Pay once, never again ✓14-day refund, no questions asked

Download & try free

Try free for 7 days, pick a plan in the app · Paddle checkout — cards, Apple Pay, PayPal, Alipay

Asked & answered.

macOS already has free dictation — why pay for Vocco?

Built-in dictation makes you toggle it on and off and leaves raw output as-is. Vocco is push-to-talk: hold one key anywhere, speak, release — the text is already pasted, punctuated, and split into clean sentences. And on mixed-language speech, the bundled engines are in a different accuracy class. Every dictation is also kept in a local, searchable history.

Which languages can I mix?

Out of the box, every language Apple's on-device recognition supports. For mixed speech, the downloadable engines (Qwen3 ASR, SenseVoice, Paraformer, Whisper) handle Chinese–English exceptionally well — our published benchmark — and cover many other languages. Everything runs offline.

Does it really work with no internet?

Yes. Transcription runs entirely on Apple silicon. The only network requests are optional: update checks and license activation.

Can it handle technical terms for coding?

Yes — engine vocabulary covers programming jargon well, and the default engine supports custom hot-words for your project's own names (coming to settings soon).

What happens when the trial ends?

Dictation pauses and your local history stays readable. Enter a license key to continue — nothing is deleted, and we don't have your card, so nothing is charged.

Is it really a one-time purchase?

Yes — $69.99 once covers the app and its updates on every Mac you own, forever. Prefer a lower entry? There's also a $6.99/month plan you can cancel anytime; access runs to the end of the paid period.

How do you handle payment?

Checkout is handled by Paddle, our merchant of record — cards, Apple Pay, PayPal, and Alipay. Your license key arrives by email within a minute.

Your voice with mixing languages, typed.

Most engines fall apart when you switch languages. Ours don't.

Prompt your agent at speaking speed.

Privacy isn't a promise. It's the architecture.

Hold. Speak. Release.

A coffee a month. Or own it forever.

Asked & answered.

Stop typing. Start talking.